Data-driven models do not rely on our knowledge of the inner workings of the process because it is often incomplete. They are estimated from data that have been collected from experiments designed to reveal the relationship between input variables, called factors, and the output measurements that characterize the performance of the process. These are the well-known Response Surface Methodology (RSM) models widely used in the analysis of experimental results. Here, we will describe a generalization of the RSM models, called Dynamic Response Methodology (DRSM) models, which also describe the time evolution of the process.
Knowldge-Driven, Data-Driven and Hybrid Models
Data-Diven Models
Knowledge-Diven Models
Knowledge-driven models are the models that are put together when one knows reasonably well the inner workings of the process. In such a case, the model consists mostly of equations that describe the material and energy balances of the processing unit of interest. We give further details about these models here.
Hybrid Models
Because Data-driven models do not use even our partial knowledge of the inner workings of the process, they do not represent the ultimate type of model we should be aiming for. Hybrid models achieve this. These consist of material and energy balances in which the terms describing the corresponding rate phenomena are estimated from process data. These hybrid models are more accurate than the pure data-driven ones and enable us to extrapolate beyond the conditions where we collected data with greater accuracy than with data-driven models. They are described here.
